Figure 1
From: Stability of similarity measurements for bipartite networks
Typical examples of the comparison between object similarities 
and 
of the MovieLens data set for fifteen similarity measurements.
When dividing the data set, we set 
, i.e. 50% ratings are divided into the first sample and the others are divided into the second one. For each calculation, we randomly select 104 pairs of objects’ similarities of two samples to show in the figure. Thus, there are 104 dots in each subplot, each represents a pair of objects. The dots would locate on the diagonal if the similarities in two samples 
. Consequently, the more stable the similarity measurement is, the more concentrated the dots would distribute around the diagonal.
